GPT 5.2 Release, Corporate Collapse in 2026, and 1.1M Job Loss | EP #215

Peter discovered this after frustration with his portfolio companies: they'd try to have AI fix their Java codebase and it would struggle, but simply saying 'scrap it, rebuild in Python' resulted in a finished product in an hour. He now generates massive volumes of code using Kimmy K2 for speed, then routes the output through GPT-5.2 to proofread for vulnerabilities because the generated code 'just flat out works' and is too voluminous for human review. He learned the hard way about spyware risks from Chinese models after Alex warned him that open-weight models can inject vulnerable code when prompted with certain topics. The protocol thus has two phases: (1) rebuild in Python with a fast model for volume, (2) proofread with a trusted safer model.

Peter and the hosts note that out of 20 major companies they work with, only three are doing even halfway the right thing. The rest are pushing the old model, assuming they can catch up later. The correct play is to find external AI-native startups (like from Y Combinator or Neo) to be development partners, invest in them, and let them build the new stack inside your organization as a tightly bounded startup. This sidesteps the talent shortage (billion-dollar signing bonuses for AI engineers) and the innovator's dilemma. The book 'Exponential Organizations' outlines the step-by-step, and Christensen's 'Innovator's Dilemma' explains why incumbents fail but not much about the prescriptive fix.

Dave points out that 20,000 highly talented people in Seattle have been cut from Microsoft and Amazon. These individuals can be retrained quickly on AI platforms and placed inside companies like State Street Bank or Walmart, where demand for AI deployment talent is desperate. The alternative — just launching software on Hacker News — fails for enterprise. The forward-deployed model creates immediate sales and deeply embeds the technology.

Dave answers a user question about how AI can lift up non-entrepreneurs. He emphasizes that historically, being a software god required deep coding skills, but with AI, it's about intuition and knowing how it applies in your industry. He advises to 'just jump in the game'. This echoes Peter's own approach of spending hours weekly with the latest models and then applying them across his dozen companies.

The ring is from Eric Mijigovski (founder of Pebble Watch), who famously used Kickstarter to validate products. This ring allows you to press a button and whisper to your ring, turning transient thoughts into structured AI input. The voice goes via Bluetooth to a locally hosted LLM on your phone, ensuring privacy. The hosts marvel at the idea of a 'button on the human body' that connects to AI, and predict next steps like injectable or edible foundation models.

The hosts applaud the pivot as brilliant: it leverages existing blade and manufacturing expertise, matches astronomical AI energy demand, and provides immediate revenue to sustain the supersonic vision. They note it's like Amazon with AWS or SpaceX with Starlink — a cash-generating business that funds the moonshot. The turbine is sexy-looking and already has $1.25 billion in backlog. The broader lesson: every company should ask what asset they have that can become a profit center in the AI ecosystem.

The hosts compare it to the movie 'Simone' and the Washington Post's decline, arguing that consumers don't crave authenticity as much as assumed — they watch whatever is interesting. AI actors will first win in TikTok and video games, then undermine traditional film budgets and audiences. The question becomes whether human actors should license their personas now before synthetics take over. OpenAI's deal with Disney to bring characters into Sora 2 shows the IP angle. Peter notes this is a 'race for neurons' — with Dunbar's number limited, synthetic personalities could fill slots in people's social circles.

Sundar's clip reveals Google's plan to launch small racks in 2027 and scale thereafter. The cooling solution: for every square meter of solar panel, only one same square meter of radiative cooling area is needed, all aluminum-based. Sun-synchronous orbits ensure constant sun exposure. The hosts discuss that failing satellites would be constantly replaced on a 3-year cycle due to chip obsolescence, so damage from solar storms or debris is less catastrophic. Fault tolerance for space-based training runs remains unsolved, but diversification across orbits could help. This shift will crowd sun-synchronous orbits, creating a new real estate gold rush regulated by the FAA.

Ben Lamb, the CEO, has no biology background but came from Hyper Giant software, and built the company from zero to $10 billion valuation in four years. Peter mentions they have 12 species in various stages, including saber-tooth tiger and mammoth. DNA older than 1.2 million years hasn't been recovered, but that still covers a wide spectrum. The process involves figuring out which gene variants produce desired traits and then splicing them into living relatives. Ben will speak at the Abundance Summit as a moonshot closing speaker, highlighting the extreme entrepreneurship angle.